《中国康复理论与实践》 ›› 2022, Vol. 28 ›› Issue (3): 335-339.doi: 10.3969/j.issn.1006-9771.2022.03.009

• 临床研究 • 上一篇    下一篇

采用多层感知器神经网络构建亚急性期缺血性脑卒中患者短期预后的预测模型

赖海芳1,顾琳2,纵亚1,牛传欣1,谢青1()   

  1. 1. 上海交通大学医学院附属瑞金医院,上海市 200025
    2. 上海市瑞金康复医院,上海市200023
  • 收稿日期:2021-10-11 修回日期:2021-11-05 出版日期:2022-03-25 发布日期:2022-03-31
  • 通讯作者: 谢青 E-mail:ruijin_xq@163.com
  • 作者简介:赖海芳(1991-),女,汉族,福建福鼎市人,硕士研究生,主治医师,主要研究方向:神经系统疾病的康复治疗。
  • 基金资助:
    上海市临床重点专科项目(shslczdzk02701);上海交通大学医学院附属瑞金医院(北部院区)研究基金项目No(2020ZY10)

Prediction of short-term outcome after subacute ischemic stroke using multiple layer perceptron neural network

LAI Haifang1,GU Lin2,ZONG Ya1,NIU Chuanxin1,XIE Qing1()   

  1. 1. Ruijin Hospital, Shanghai Jiaotong University School of Medicine, Shanghai 200025, China
    2. Shanghai Ruijin Rehabilitation Hospital, Shanghai 200023, China
  • Received:2021-10-11 Revised:2021-11-05 Published:2022-03-25 Online:2022-03-31
  • Contact: XIE Qing E-mail:ruijin_xq@163.com
  • Supported by:
    Shanghai Municipal Key Clinical Specialty No(shslczdzk02701);Ruijin Hospital, Shanghai Jiaotong University School of Medicine (Northern Campus) Research Fund No(2020ZY10)

摘要: 目的 采用多层感知器(MLP)神经网络构建亚急性期缺血性脑卒中(CIS)短期预后的预测模型。方法 2019年1月至2021年9月,上海市瑞金康复医院康复科再住院缺血性脑卒中患者60例,采集首次住院时(病程< 30 d)的临床相关信息,根据首次入院3个月后改良Rankin量表评分,判断患者短期预后。采用单因素分析筛选与短期预后相关的危险因素,分别采用常规多因素Logistic回归和MLP建立预测模型,计算两种模型的预测准确率,采用接受者操作特征(ROC)曲线评估预测效应。结果 多因素Logistic回归模型预测准确率73.3%,ROC曲线下面积0.851;MLP模型预测准确率88.9%,ROC曲线下面积0.930。结论 采用MLP模型能更好预测亚急性期缺血性脑卒中的短期预后。

关键词: 缺血性脑卒中, 亚急性期, 多因素Logistic回归, 多层感知器, 神经网络, 预测, 短期结局

Abstract: Objective To establish a predictive model using multiple layer perceptron (MLP) for short-term outcome after subacute ischemic stroke.Methods From January, 2019 to September, 2021, 60 readmission-inpatients in Department of Rehabilitation, Ruijin Hospital, Shanghai Jiaotong University School of Medicine were collected the clinical features of first admission (less than 30 days after attack), and the outcomes were assessed with modified Rankin Scale (MRS) three months after the first admission. The risk factors were screened with single factor analysis, and the short-term outcome predictive models were established with multi-factor Logistic regression and MLP. The predictive accuracy of both models was calculated, and the predictive effects were compared with Receiver Operating Characteristic (ROC) curve.Results For multi-factor Logistic regression, the predictive accuracy was 73.3%, and the area under ROC curve was 0.851. For MLP, the predictive accuracy was 88.9%, and the area under the ROC curve was 0.930.Conclusion The prediction of short-term outcome after subacute ischemic stroke can be done with MLP model.

Key words: ischemic stroke, subacute, multi-factor Logistic regression, multiple layer perceptron, neural network, prediction, short-term outcome